The Agent Skills Directory

[SAFE]: The skill defines a benign process for prompt engineering, using structured assessment and clear instructions to improve the quality of AI responses without requesting sensitive permissions.
[COMMAND_EXECUTION]: The skill instructs the agent to manage a local prompt library by creating a directory (.ai/) and writing content to a file (.ai/PROMPT.md). These operations are confined to the local project workspace and do not involve system-level changes or privilege escalation.
[PROMPT_INJECTION]: The skill contains no instructions designed to override agent safety protocols or exfiltrate data. While it processes untrusted user input (the prompts to be refined), the output is strictly markdown content meant for local storage and does not present a path for execution or exfiltration.

refine-prompt